model-writtenrun 6309 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed replacing lengthy lists of acceptable relation kinds with an error describing the failed operation and a shared detail identifying the actual relation kind, also including RELKIND_HAS_STORAGE cleanups #1.
- Robert Haas requested clearer operation names and consistent trigger/rule wording, while Tom Lane proposed details explaining that the operation is unsupported for the entire relation kind; Andrew Dunstan supported that formulation #2 #3 #4.
- Michael Paquier suggested retaining relation names in details, separating storage cleanups, and simplifying schema-change error construction, but Alvaro Herrera favored Tom's generic detail because the primary error already names the relation and the restriction applies to its whole kind #5 #6.
- Alvaro Herrera proposed spelling out ALTER subcommands, and discussion established that the message should identify the attempted action without assuming ALTER TABLE, since the same code also handles ALTER VIEW #7 #8 #9 #11 #12.
- After more than a year, Peter returned with a revision adopting errdetail_relkind_not_supported(), adding ALTER subcommand descriptions, extending contrib coverage, and omitting the earlier standalone storage cleanups in heap.c, pgstat.c, and dbsize.c #13.
- Michael reviewed the revision and requested a translator comment and more precise DETACH PARTITION wording; Peter accepted those edits but retained schema-change hints and the categorical 'cannot have triggers/rules' wording, which Alvaro also supported #14 #15 #16.
- Peter committed the revised patch with only the translator comment and DETACH PARTITION wording adjustment as 2ed532ee8c, then announced completion #13 #17.
- While reviewing logical decoding of sequences, Peter identified publication and subscription errors that should use the new convention and submitted a separate follow-up covering those sites #18.
- Michael approved the follow-up and suggested examining init_sequence(), RangeVarCallbackForAlterRelation(), and ExecGrant_Relation(); the submitted publication and subscription changes landed unchanged as 81d5995b4b #18 #19.
- Both submitted changes had landed by the final reply, where Peter explained that 'not a sequence' remains appropriate for an explicit ALTER SEQUENCE command and left the other suggested sites for further investigation #20.